< Back to previous page

Project

Algorithm-Hardware Co-Design for Adaptive Embedded Anomaly Detection

Intelligent machines are the key to the 4th industrial revolution. These intelligent machines are equipped with smart sensors and cyber-physical systems. These enable the machines to take decisions on their own and to perform their tasks as autonomously as possible. These automated machines dramatically improve the quality and efficiency of production processes, causing their reliability to become increasingly more important to guarantee safety and limit costly machine downtime. By monitoring the health of a machine continuously by using the available sensors, maintenance can be scheduled more efficiently to improve reliability. To enable this, a continuous stream of sensor data is required, which is typically analyzed in the cloud. This continuous stream of data towards to cloud puts a strong pressure on the network capacity. In this research we apply anomaly detection to model the normal behavior of a machine. By introducing prior knowledge, the amount of required training data can be reduced. When an error occurs, the machine will deviate from its modeled normal behavior, which can trigger the schedule of maintenance. This research will also cover energy efficient hardware implementations for both inference as online learning of the model. This will enable to do the required computation close to the machines, which is beneficial for the network infrastructure and ease of use of the final product.

Date:1 Apr 2019 →  15 Mar 2023
Keywords:Artificial Intelligence, Machine Learning, Anomaly Detection
Disciplines:Machine learning and decision making
Project type:PhD project